Messages in this thread | | | Date | Sun, 4 Jan 2009 14:11:59 +0300 | From | Cyrill Gorcunov <> | Subject | Re: early exception error |
| |
[Jiri Slaby - Sun, Jan 04, 2009 at 01:24:03AM +0100] | On 01/03/2009 10:24 PM, Cyrill Gorcunov wrote: | > [Cyrill Gorcunov - Sat, Jan 03, 2009 at 10:03:16PM +0300] | > | (list restored) | > | | > | [david@lang.hm - Sat, Jan 03, 2009 at 11:19:00AM -0800] | > | ... | > | >>> | > | >>> two new screenshots at http://linux.lang.hm/linux | > | >>> | > | >>> 36 is a boot with just earlyprintk=vga | > | >>> 37 is a boot with numa=noacpi | > | >>> I also put the vmlinux file there, I'll put the System.map and config | > | >>> there later (I did enable kernel_debug on this build as well) | > | >>> | > | >>> David Lang | > | >>> | > | >> | > | >> David, I can't find vmlinux neither .config? | > | >> Maybe they have hidden attribute? | > | | > | ok, according to failing address we've a BUG_ON | > | triggered | > | | > | --- | > | (gdb) l *0xffffffff8096452a | > | 0xffffffff8096452a is in alloc_bootmem_core (mm/bootmem.c:442). | > | 437 unsigned long fallback = 0; | > | 438 unsigned long min, max, start, sidx, midx, step; | > | 439 | > | 440 BUG_ON(!size); | > | 441 BUG_ON(align & (align - 1)); | > | 442 BUG_ON(limit && goal + size > limit); | > | 443 | > | 444 if (!bdata->node_bootmem_map) | > | 445 return NULL; | > | 446 | > | (gdb) | > | --- | > | | > | so we're in attempt to overrun 'limit'. | > | Hmm... | > | | > | - Cyrill - | > | > Hardly possible that we trigger BUG here since I don't | > see BUG: on the photo. Investigating. | | Hint: line 442 in 2.6.28 is | if (!bdata->node_bootmem_map) | ;) | | It's: | 0xffffffff8096452a <alloc_bootmem_core+69>: cmpq $0x0,0x10(%rbp) | and hence cr2 is 10. | | node_data[nid] is NULL... But both of them are set up. Maybe too high nid (and | pnum in sparse_init)? |
It seems to be true!
What is worse we have a number of __nr_to_section users which don't check for NULL returned and secondly
static inline struct mem_section *__nr_to_section(unsigned long nr) { if (!mem_section[SECTION_NR_TO_ROOT(nr)]) return NULL; return &mem_section[SECTION_NR_TO_ROOT(nr)][nr & SECTION_ROOT_MASK]; }
SECTION_NR_TO_ROOT is not modulo operation so we could run out of mem_section[NR_SECTION_ROOTS].
David I'll cook some testing patch shortly. Many thanks to Jiri!
- Cyrill -
| |